提出了一种基于听觉事件检测的汉语声韵母切分方法。该方法首先使用耳蜗滤波器组对语音进行滤波,然后在每个频带上检测对应于能量突变的听觉事件,最后在不同频率范围对听觉事件进行融合以确定声韵母边界。实验结果表明,对8 kHz采样的干净语音切分准确率可达到88.9%;信噪比10 dB的语音切分准确率可达到82.9%以上。 This paper proposes a segmentation method of Chinese vowel based on auditory event detection. The method first filters speech using a cochlear filter bank, then detects auditory events corresponding to abrupt changes of energy on each frequency band, and finally fuses auditory events in different frequency ranges to determine the boundary of the vowels. Experimental results show that the accuracy of the clean speech segmentation for 8 kHz sampling can reach 88.9%, and the speech segmentation accuracy of 10 dB SNR can reach more than 82.9%.